This week’s reading list is about costs that turn up in a different account from the one you were watching. The safest asset in the world may be where the next crisis starts, because the country that has always stabilised the system may no longer be willing to. A war fought over there arrives at the petrol pump over here, and mostly through refining capacity rather than the crude price everyone quotes. Artificial intelligence shows up in the labour market not as redundancies but as jobs that were quietly never advertised, and the bill lands on people trying to enter the workforce for the first time. A US$750 billion data centre buildout arrives in portfolios through equities, credit, property and infrastructure all at once, which means the concentration is real even when no single allocation looks concentrated. A US$17 billion penalty lands not on the company that paid it but on the competitors who now have to match a standard the settlement wrote for them. And eighty five years of compounding, we are reminded this week, was paid for in a currency that never appeared on any balance sheet. The useful lesson for investors is that the ledger balances eventually, but rarely in the line you were reading. The best investors are not only asking what something costs. They are asking who ends up paying, and when the invoice turns up.

📚 The ‘reverse Kindleberger Trap’: reasons to worry about the next financial crisis: David Lubin at Chatham House on the uncomfortable question of who stabilises the system next time. Charles Kindleberger argued the 1930s Depression ran as long and as deep as it did because Britain was unable to act as a responsible hegemon and the United States was unwilling. Lubin’s point is that the trap may now run in reverse: China has the ambition but not the financial capacity, while the United States retains the capacity and may lose the willingness. The contrast with 2008 is stark, when the Federal Reserve pushed dollar liquidity out through swap lines to fourteen central banks and bought mortgage securities, more than half of which had originated with foreign firms. The line that matters most for anyone holding unhedged offshore assets is the last one: a US crisis is now more likely to produce a capital outflow than an inflow, with the dollar weakening substantially rather than rallying. (Chatham House)

📚 Labor Day on Track to Set Record at the Pump: The AAA numbers, straight from the source. The US national average has reached US$4.14 a gallon, the highest ever recorded for this point in the year, with crude sitting around US$90 and volatility through the Strait of Hormuz keeping it there. Worth pairing with Paul Krugman’s piece on the same shock, which makes the more useful analytical point: diesel and petrol have risen considerably more than crude, because the binding constraint right now is refining capacity rather than barrels in the ground. Ukrainian strikes on Russian refining assets are a large part of that. For investors with energy exposure, the distinction between a crude price story and a refining margin story is the whole thing. (AAA) see also Imperialist Delusions and the Price of Fuel (Paul Krugman)

📚 Job Postings Show Early Signs of AI Automation Impact: Samuel Dodini and Tucker Smith at the Dallas Fed with the cleanest evidence yet that the AI labour effect has arrived, and that it is showing up in hiring rather than firing. Two thirds of Texas firms surveyed in May reported using AI, up from 40 per cent two years earlier. Postings for occupations most exposed to automation fell about 5 per cent by the end of 2023 and roughly 8 per cent by early 2025, with surviving incumbent firms accounting for the decline rather than new AI native competitors. Across the whole state, the estimated drag on total online job postings was 1.8 per cent in 2024 and 2.6 per cent in 2025. The distributional finding is the sharp one: because online postings skew toward roles requiring little experience, the cost falls hardest on new graduates and career changers, which is a labour market problem that will not appear in an unemployment print for some time. (Dallas Fed)

📚 Data Center-Related Investments Available Across Most Asset Classes: Bailey McCann with the number that frames the theme, US$580 billion invested in data centres in 2025 according to Colliers, about to be dwarfed by the US$750 billion hyperscalers plan to spend this year. The genuinely useful part is not the size but the shape. Depending on which slice you own, land, power generation, transmission, construction, cooling or the decommissioning of old capacity, the exposure sits in your equities, your credit, your property or your infrastructure allocation. Mercer’s Alan Synnott argues for total portfolio analysis so investors can work out what they already own before adding more. Eric Lang at the Texas teachers’ fund puts it more bluntly, saying they now ask of every investment whether it is going to be interrupted by AI. Note also the copper observation, and CalPERS on the risk of a state level legislative reaction to the buildout. (Chief Investment Officer)

📚 Little Tobacco Moment: Scott Galloway on the settlement between Meta and 47 states plus the District of Columbia, resolving claims around addictive algorithms aimed at minors, youth privacy and child endangerment. Up to US$17.1 billion in penalties, no admission of wrongdoing. His comparison is the instructive one: the 1998 Tobacco Master Settlement was US$206 billion, or roughly US$422 billion in today’s money, against a far smaller industry. Set against Meta’s US$200 billion of 2025 revenue and a US$1.4 trillion market capitalisation, the number does not function as a deterrent. The part investors should sit with is what happens next. Compliance obligations become a moat against new entrants, and US$5 billion of the penalty is conditioned on competitors agreeing to similar terms, which turns a punishment into a competitive weapon. (Scott Galloway)

📚 The Cost of Being Warren Buffett: Vishal Khandelwal, marking Buffett’s 96th birthday on 30 August, on why nobody has replicated a man who explained himself more openly than any investor in history. His answer is not intelligence. It is that the ideas were never the hard part. He is good on the structural advantage most people gloss over, that insurance float cannot be redeemed at the bottom the way a fund’s capital can, so Buffett engineered away the single mechanism that destroys good managers. He is better still on what the temperament cost, drawing on Schroeder for the parts of the story that do not fit on a motivational poster, and on the honest question of how much of the record belongs to the era and the country rather than the man. A good piece to read slowly, and an unusually clear one on the difference between what you can copy and what you cannot. (Safal Niveshak)

📚 I Rented a Car, and Within Hours, My Driver’s License Was for Sale: Dan Goodin discovers his own licence among more than 153 million scans offered on a dark web marketplace, alongside millions of ID cards, travel documents and medical cards, complete with infrared and ultraviolet imaging. The trail leads to an identity verification provider, which is to say a company whose entire product is preventing exactly this. The FBI is investigating. There is a specific investment point buried in the story and it is not about cybersecurity stocks: every business that has quietly outsourced identity verification to a third party has taken on a liability it does not control and probably has not priced. (Ars Technica)

📚 Humans Did Not Invent Art. It Was the Other Way Around: J F Martel on the Lascaux caves, found in 1940 by four teenagers chasing a rumour about secret passageways, and on the strange fact that recognisable human figures do not appear in cave art until roughly eight thousand years after those paintings were made. His argument is that art did not express a humanity that already existed, it produced one. There is a sharp aside near the end on why artificial intelligence can only ever generate artifice, because it works from things already made rather than things in the making. No portfolio application at all. We include it anyway. (Aeon)